一种基于多传感器的红外图像正则化超分辨率算法

Infrared image regularization super-resolution algorithm based on multi-sensor

提出一种红外图像多传感器超分辨率重建算法。算法存在两个关键点:一是有效利用两类图像的相关性;二是针对红外图像的特点利用其自身信息构造正则化模型。采用相位一致性算法提取可见光图像边缘,利用此边缘信息对正则化模型加权,以充分利用可见光和红外图像的相关性;将一阶梯度锐化算子引入总广义变分模型,构成针对红外图像特点的正则化模型;最后采用一阶主-对偶优化算法求得加权后模型的最优解。实验表明,本文算法可获得边缘清晰的重建结果,并且有效抑制噪声,在主观视觉效果和客观评价指标方面均优于其他算法。

It＇s necessary to improve the resolution of infrared image because the tow-resolution image can- not meet the demand of many applications. Traditional approaches reconstruct infrared image merely from low-resolution infrared image, which deliver limited results. High-resolution visible image, by con- trast,can be easily obtained with a CCD camera and has a strong correlation with the infrared image, from which we can increase the resolution of infrared image by utilizing the information of visible image. This paper presents a new infrared image resolution improvement framework based on multi-sensor. The proposed method consists of two key points. The first one is that the correlation between infrared and visible images should be used efficiently;the second one is that the regularization model should be suit- able for infrared image super-resolution. We use phase congruency to extract the edges of visible image, and the edges are then combined with a regularization model, which utilizes the correlation sufficiently. In addition,the regularization model is built by first-order graduate operator and total generalized variation regularization, which is applicable to the reconstruction of infrared image. Finally, this method infers the super-resolved infrared image with a first-order primal-dual optimization scheme. Experimental results demonstrate that the proposed method can obtain clear results and suppress noise effectively. When com- pared with other methods,the proposed algorithm is superior in terms of subjective and objective quali- ties.